Artificial neural networks (ANN) are intelligent systems that are capable to simulate extremely complex functions [1, 2, 4]. ANN-non-linear models can be used with dependent data sets and subsequently be trained on examples [3, 6]. The user just needs to collect representative data and apply learning algorithms for the automatically recognizing of the data structure. The detection of variable results in various studies indicates the need for an effective mathematical model. For selecting the model and interpret the results of the work of ANN it is worthwhile to involve the knowledge of expert-researchers in a specific subject area.

Purpose.

The aim of the study was to develop a system for optimization of evaluation of thickening of the vascular wall using ANN.

Materials and methods. 

Carotid ultrasound was performed on 242 patients (with average age 67, 03 ± 7, 3 years) by duplex scanning on ultrasonic system Voluson 530 DMT (Kretztechnik, Austria) with linear transducer 5, 5 - 7, 0 MHz. IMT CCA was determined at the distal point of the distal centimeter CCA [5]. Patients were divided in 4 groups depending of the thickness of the intima-media-thickness of  common carotid artery (IMT CCA): 1) normal (IMT <1,0 mm) (n=97); 2) IMT 1,1-1,3 mm (n=51); 3) 1,4 mm<IMT<2,0 mm (n=77); 4) IMT > 2 mm (n=57).

We used a mathematical classification to optimize the assessment of carotid arteriosclerosis. ANNs of forward and backward propagation were composed of neurons with the input, hidden and output layers. In our experiment, the output vectors are classified into four categories through four neurons. Parameters related to the input layer indicate the degree of change of the IMT CCA and determine the values of the neurons of the output layer. Randomized use of parameters to neurons in the input layer obtained by examining all patients activates ANN. The values of the neuron at the output of the neural network (calculated diagnosis) were compared with real information on the degree of IMT, determined by experienced physicians, and the differences between them were calculated as an error. To check the validity of the developed models, approximately 50% of the database was used. This tactic is used to optimize the model.

Results.

In this research, we selected indicators of the grade of change in the IMT CCA. In order to facilitate the further processing of data from a large number of input parameters, we selected the most important (informative) indicators (signs) that affect the degree of increase in IMT. The applied ANN model reduced the effect of input variables, which do not have a significant influence on the output. 14 of the 21 parameters had the greatest impact on the degree of IMT CCA.

The research showed that the important role in the thickening of the arterial wall is provided by the levels of initial diastolic pressure (weight 0.43), systolic blood pressure (0.53), mean BP (0.54), age (0.51). The revealed parameters are the most informative, simple, non-invasive markers of thickening of the vascular wall.  The chosen parameters were applied to the ANN model as independent variables and was realized their training. An ANN can be trained according to the nature of the problem and medical expectations. In addition to the possibilities of implementing a diagnostic expert system in real time, a diagnosis can be established more accurately by increasing of the diversity and the number of studied parameters.

---

1. Ignatiev N.A. Extracting explicit knowledge from different types of data using neural networks // Computational technologies. - Novosibirsk, 2003.- V.8- No. 2.-P.69-73.

2. Ignatiev NA, Madrakhimov Sh.F. About some ways of increase of a transparency of neural networks // Computational technologies. - Novosibirsk, 2003.-V.8.- No.6.- P.31-37.

3. Mobley B.A., Schecter E., Moore W. E., McKee P.A., Eichner J.E. Predictions of coronary artery stenosis by artificial neural network//Artif. Intell. Med. 2000.- No.18.- P.187-203.

4. Rozikhodjaeva G. A., Ikramova Z.T.,Rozikhodjaeva D.A. Neural network system in the building information models of degree of changes of vascular wall in patients with carotid atherosclerosis// Philosophical Problems of Information Technologies and Cyberspace. 2012.- No.1.- P. 73-80.

5. Pignoli P., Tremoli E., Poli A. et al. Intimal plus medial thickness of the arterial wall: a direct measurement with ultrasound imaging. // Circulation. 1986.- No.74. – P.1399-1406.

6. Serhatl oglu S., Hardalac F., Guler I . Classification of transcranial Doppler signals using artificial neural network// J. Med. Systems. 2003.- V.27.- No.2.- P. 205-214.